ChatGPT 会“幻觉”吗？探讨人工智能的局限性
我一直在思考 ChatGPT 回答的准确性。有时它看起来知识非常丰富，但有时又会提供不太准确、甚至完全编造的信息。ChatGPT 是否会出现“幻觉”，即生成虚假或捏造的信息？这种行为发生的频率如何？其成因又是什么？期待听取有此经历的专家和用户的分享。
Grayson Newton
March 9, 2026 at 03:34 PM
我一直在思考 ChatGPT 回答的准确性。有时它看起来知识非常丰富，但有时又会提供不太准确、甚至完全编造的信息。ChatGPT 是否会出现“幻觉”，即生成虚假或捏造的信息？这种行为发生的频率如何？其成因又是什么？期待听取有此经历的专家和用户的分享。
添加评论
评论 (3)
The hallucination issue arises because the model predicts text based on probability, not truth. So if it hasn't seen accurate info or if prompts are ambiguous, it might 'guess' wrong.
I've noticed ChatGPT sometimes fabricates sources or quotes that don't exist. It's a limitation to be aware of when using it for research.
Yes, ChatGPT can hallucinate. It generates responses based on patterns in data, but it doesn't have true understanding, so it sometimes produces plausible-sounding but incorrect information.